Build Wispr Flow voice inbox and capture handoff

[WalksWithASwagger]

Context

Wispr Flow is now the user's live dictation layer. WhisperForge needs a first-class inbox for pasted dictation, uploaded audio, imported transcripts, and raw notes so capture can feed durable knowledge work instead of being trapped in ad hoc text fields.

Source: docs/WHISPERFORGE-MASTER-PLAN-2026-05-18.md (wf-voice-inbox, Phase 1).

Linear: BC-221.

Acceptance Criteria

• Add a capture inbox model with source type, created time, title, status, and linked run identifiers.
• Add a fast paste/import path optimized for Wispr Flow output and other already-transcribed text.
• Normalize captures before pipeline execution so audio, pasted transcript, and imported note flows share downstream metadata.
• Persist capture metadata into local run artifacts and markdown/Notion export metadata where applicable.
• Document the capture workflow in README or the roadmap docs.

Tests/Evals

• Unit tests for capture metadata serialization and normalization.
• Rendered UI smoke coverage if a visible inbox surface is added.
• Manual fixture using pasted Wispr Flow-style text.

Verification

• make test
• venv/bin/python tests/ui_smoke.py
• git diff --check

Agent Instructions

Keep this as a capture layer, not a live dictation clone. Match existing Streamlit UI patterns and store artifacts under the existing cache/run architecture.

Out of Scope

Do not build a system-wide dictation overlay, browser extension, or mobile app in this issue.

[WalksWithASwagger]

PR #26 is ready for review: https://github.com/WalksWithASwagger/spektorAI/pull/26\n\nThis slice adds durable Wispr Flow/paste/upload capture records, links captures to run metadata and source receipts, and exposes the Capture inbox. Verified with make test, make eval-fixture, ui_smoke, make smoke, local Playwright UI checks, git diff --check, and ops/roadmap/features.json validation.